Dexcom tells Assembly its AI 'health coach' is tightly guarded and not a dosing tool

Ginny Hu, senior director for regulatory affairs at Dexcom, described how the company incorporated generative‑AI features into its Stell O biosensor mobile app and how the device fits FDA’s total product lifecycle framework. She emphasized risk management across design, cybersecurity and post‑market surveillance, and explained concrete guardrails that prohibit clinical dosing advice: "That is one of the things that we we've, made sure that we put in the safety guardrail...the AI is not able to give insulin dosing advice," she said.

Hu said Dexcom worked closely with FDA on software verification and validation, cybersecurity controls (ISO and SOC standards) and ongoing surveillance of deployed features, and noted that the company was selected for the FDA Tempo pilot to explore digital health and aggregate clinical‑decision support. She said the AI coach provides contextualized insight from dense CGM data (288 points per day) but defers diagnostic or dosing questions back to clinicians when necessary.

Why it matters: As regulated medical‑device manufacturers add AI UX features, Dexcom’s presentation shows a pathway for deploying consumer‑facing AI while retaining clinical safeguards. The company’s approach foregrounds continuous monitoring and human oversight as mechanisms to reduce risk.

Next steps: Lawmakers asked how consumer and aggregate data can be used to improve products and how to protect patient privacy; Hu flagged ongoing collaboration with FDA and pilot programs to refine frameworks for clinical‑decision support.